SourceMux exhibits significant definition quality gaps. While all 12 tools have names and descriptions, most descriptions are terse (10-50 chars, well below the 194-char baseline). Input schemas are minimal: most tools accept only 1-2 parameters with basic type info but lack detailed parameter descriptions, constraints, or output schema documentation. No tool declares enums, ranges, or validation rules. Error handling is absent from visible code. The server exposes multiple external API integrations (Exa, Tavily, Firecrawl, TinyFish, Jina) but provides no guidance on which tool to use when, no dependency hints, and no recovery paths. Tool composition is weak: search, exa_search_advanced, docs_search, and search_planning all perform search-like operations with no clear differentiation. No tool returns structured output schemas or pagination info. This is a typical community server in the 40-50 range.

• Expand tool descriptions to 100-200 chars. For each search-like tool, explicitly state: 'Use this when you need [specific data type] from [specific source]. Returns [output format]. Faster/slower than alternatives: [comparison].' Example: 'Use exa_search_advanced for academic papers and research articles with metadata. Returns structured results with URLs, snippets, and publication dates. Slower than web search but higher quality for research.'
• Add parameter descriptions with constraints. For 'query': 'Search query (1-500 chars, natural language or keywords). Supports boolean operators (AND, OR, NOT) and quoted phrases.' For 'url': 'Full URL including protocol (http:// or https://). Must be publicly accessible.'
• Document output schemas for every tool. Use JSON Schema or prose. Example for search(): 'Returns array of {url: string, title: string, snippet: string, source: string}. Max 20 results per call. Use pagination cursor if available.'
• Consolidate overlapping search tools. Merge search, docs_search, exa_search_advanced, and search_planning into a single search_web tool with a 'source' enum parameter (web|docs|academic|planning). This reduces LLM confusion and simplifies composition.
• Add error handling with recovery hints. Wrap external API calls with try-catch. Return structured errors: '{error: string, code: string, retry: boolean, suggestion: string}'. Example: '{error: "URL unreachable", code: "FETCH_TIMEOUT", retry: true, suggestion: "Try fetch() with a different URL or use search() to find an alternative source."}'